spotbugs: fix PZLA, SI, DB_DUP, MC_OVERRIDABLE — 4 small residual findings

[Vest]

What

Closes four small SpotBugs findings that no other open PR covers.

Rule	Site	Fix
SI_INSTANCE_BEFORE_FINALS_ASSIGNED	Region.java:41	Reorder static fields; drop redundant = 0
DB_DUPLICATE_SWITCH_CLAUSES	HitPointFacet.java:94	Name HP_STANDARD explicitly; make default an explicit warn-and-fall-back
PZLA_PREFER_ZERO_LENGTH_ARRAYS	FactSetParser.java:137	Suppress with rationale (null = errored)
MC_OVERRIDABLE_METHOD_CALL_IN_CLONE	CDOMObject.java:1061	Suppress with rationale (PCClass override is required)

Why each

SI / Region.java — public static final Region NONE = new Region(...) ran the constructor (which does ordinal = ordinalCount++) before the explicit private static int ordinalCount = 0 initializer below it. Today it works because Java zero-inits the int, but the explicit = 0 re-zeros the counter after NONE's construction. If anyone added a second public static final Region X = new Region(...) constant later, it would collide on ordinal 0. Real latent bug. Fix: declare the counter/map first, drop the redundant = 0, then NONE.

DB_DUP / HitPointFacet.rollHP — the HP_STANDARD case and default case had identical bodies:

case Constants.HP_STANDARD -> roll = Math.abs(RandomUtil.getRandomInt((max - min) + 1)) + min;
default                    -> roll = Math.abs(RandomUtil.getRandomInt((max - min) + 1)) + min;

The two cases mean different things, though — HP_STANDARD is the named, documented primary mode (it's the default value loaded by SettingsHandler.getPCGenOption("hpRollMethod", Constants.HP_STANDARD)), and default is the safety net for unknown values (corrupted config). Collapsing them either way loses information. Fix: keep HP_STANDARD as its own named case, and turn default into an explicit warn-and-fall-back so SpotBugs no longer sees a duplicate body:

case Constants.HP_STANDARD -> roll = Math.abs(RandomUtil.getRandomInt((max - min) + 1)) + min;
default -> {
    Logging.errorPrint("Unknown HP roll method " + SettingsHandler.getHPRollMethod()
        + "; falling back to HP_STANDARD");
    roll = Math.abs(RandomUtil.getRandomInt((max - min) + 1)) + min;
}

PZLA / FactSetParser.unparse — return null is a deliberate sentinel: it distinguishes "this token errored out (already called addWriteMessage)" from "this token has no output for this object" (which returns new String[0]). The codebase uses this same idiom in TemplateFeatToken.unparse and other parsers. Suppressing with rationale is correct; the alternative would change semantics callers rely on.

MC_OVERRIDABLE / CDOMObject.clone — calls clone.ownBonuses(clone), and ownBonuses is public/overridable. PCClass.ownBonuses (PCClass.java:1418) overrides it to recursively re-own bonuses on its PCClassLevel children — making it final would break that. The override only reads the clone's BONUS list, which super.clone() populated before the call, so the "half-cloned" warning doesn't apply in practice. Suppress with rationale.

Verification

• ./gradlew compileJava compileTestJava compileItestJava compileSlowtestJava — clean
• ./gradlew test — all unit tests pass
• ./gradlew clean spotbugsMain — the four target rules each report 0:
  • SI_INSTANCE_BEFORE_FINALS_ASSIGNED: 0
  • DB_DUPLICATE_SWITCH_CLAUSES: 0
  • PZLA_PREFER_ZERO_LENGTH_ARRAYS: 0
  • MC_OVERRIDABLE_METHOD_CALL_IN_CLONE: 0
• SpotBugs total: 77 → 71 (-6; the Region reorder also cleared one EQ_COMPARETO_USE_OBJECT_EQUALS and one CT_CONSTRUCTOR_THROW as a side-effect).

Files

code/src/java/pcgen/cdom/base/CDOMObject.java
code/src/java/pcgen/cdom/content/factset/FactSetParser.java
code/src/java/pcgen/cdom/enumeration/Region.java
code/src/java/pcgen/cdom/facet/HitPointFacet.java